(2012) Evolution of the Human Biometric Sensor Interaction model

BIOMETRICS LABBiometric Standards, Performance and Assurance LaboratoryDepartment of Technology, Leadership and Innovation

EVOLUTION OF THE HBSI MODEL

STEPHEN ELLIOTTIPBC 2012 CONFERENCE PRESENTATION14:25-14:50 - 3/8/2012

CONTRIBUTORS TO THE PRESENTATION

• Michael Brockly• Kevin O’Connor• Carl Dunkelberger• Tyler Veegh• Thomas Cimino• Amanda Simpson• Rob Pingry• Brent Shuler

• Jacob Hasslegren• Rob Larsen• Chris Clouser• Dan Vander Wall• Todd Walters• Craig Hebda• Weng Kwong

Chan• Tony Fuji

PRESENTATION

• Evolution of the model• HBSI v3.0• Future roadmap

DEVELOPMENT OF THE MODEL

• The HBSI model is concerned with the data collection portion of the biometric model– Consistent and repeatable presentation to

the sensor

HBSI MODEL

Conceptual model for HBSI

Human

Biometric SystemSensor

Ergonomics

Usability

Image Quality

Human-BiometricSensor Interaction (HBSI)

UNDERLYING MODEL

MODALITY TESTING AND HBSI

Year Hand Finger Iris Face DSV

2004 Age Mobile iris Illumination Different devices

2005 Co-Rec

2006 Height /Placement

2007 Habituation Force

2008 Gender

2009 Initial HBSI Calc Force Training

2010 Fixed iris

2011 Gender Device (different sensors)

2012 Hand alignment ForceFinger interactions /

Kinect

HBSITraining /

Kinect

Detractors Forgery

2012 Interaction Age Interaction Age Interaction Age Interaction Age Interaction Age

MODEL DEVELOPMENT - V1

MODEL DEVELOPMENT - V2

INCLUSION OF OTHER MODELS

• General Biometric Model• Operation Times Model (Lazarick,

Kukula, et.al)

Fuji, et. al (2011)

HBSI METRICS V2

Metrics created and validated for:• Iris• Fingerprint• Signature Verification

HBSI METRICS V2

Metrics created and validated for:• Iris• Fingerprint• Signature Verification

Video record the environment from different angles in order to watch the

subject and to classify their presentation

-typically 3 video angles and operator

screen + (audio sometimes)

HBSI METRICS V2

Metrics created and validated for:• Iris• Fingerprint (different sensors)• Signature Verification

Record the environment (video

and sometimes audio) from different

angles in order to watch the subject

and to classify their presentation

-typically 3 video angles and operator

screen

BIOMETRICS LABBiometric Standards, Performance and Assurance LaboratoryDepartment of Technology, Leadership and Innovation

HBSI MODEL 3.0

UNDERLYING MODEL

UNDERLYING MODEL EXAMPLES

More actors:• Subject (typically

the biometric donor)• Operator • Other people in the

environment

UNDERLYING MODEL EXAMPLES

Additional variables: Subject conditions• moisture, • elasticity, • oilinessCollected conditions such as:•

temperature • humidity

EXAMPLES: HAND

GEOMETRY ACCESSIBILITY

Biometric DonorBiometric Sensor operator

Climate Device

Test Administrator

Training

EXAMPLES: HAND

GEOMETRY ACCESSIBILITY

Biometric DonorBiometric Sensor operator

Climate Device

Test Administrator

Training

EXAMPLE METRIC

CALCULATIONS

Biometric DonorBiometric Sensor operator

Climate Device

Test Administrator

Training

DETERMINATION OF ERRORS V1

• Process:– Recorded in real time as the study is

underway– Interactions are coded– Metrics of the evaluation model are

completed– Interaction errors are classified as HBSI

terms

CURRENT WORK

• Generation of the errors is time consuming• We notice other potential errors

• Contribution of the operator to the error• Contribution of the test administrator to the error

• HBSI workflow• Semi-automatic coding of the model using Kinect

• New work • Accessibility study – hearing and sight impaired (started Jan 2012)

• Contribution of cost to the model (started Jan 2011)• Examining the role of the impostor (thinking …. As this model

only has been rested in a “genuine” environment)• Development of products that can help improve interactions

ASSIGNING A COST MODEL

• Identify interaction issues• Classify where these errors are occurring

– what is causing this• Assign a cost to “retry” for example –

based on:– Poor interaction– Sensor feedback– Operator not paying attention

• Assess the impact on fixing this error

BIOMETRICS LABBiometric Standards, Performance and Assurance LaboratoryDepartment of Technology, Leadership and Innovation

KINECT

CRAIG HEBDA | ROB PINGRY | WENG KWONG CHAN |BRENT SHULER

MOTIVATION

• Video coding is time consuming• Inter-rater reliability

– Requires good robust definitions

DEFINING MOVEMENTJake Hasselgren (2011)

Slouched: Subject is not standing up straight during fingerprint scan.

Head Movement: Subject’s head is not still during fingerprint scan.

Body Movement: Subject’s body is not still during fingerprint scan.

Upright: Subject is standing up straight during fingerprint scan.

Labored Walking: Subject has bag or other item on shoulder when approaching device

Pivoting Palm: Subject’s hand pivots on edge of device

Rocking Fingers: Subjects fingers rock from one finger to the next when hand is placed on device

Slapping Hand: Subject slaps hand on to the device

Angled Fingers: Subjects fingers are at an angle other then 90 degrees from edge of the device

MICROSOFT ® SDK INTERFACE

MICROSOFT ® KINECT™ SKELETAL TRACKING SYSTEM

Source: http://msdn.microsoft.com/en-us/library/hh438998.aspx

Source: http://www.genbetadev.com/herramientas/disponible-el-sdk-de-kinect-para-desarrollar-nuestras-propias-aplicaciones-usando-los-sensores

SLOUCHING

• Dictionary Definition• Slouching-A gait or posture characterized by an

ungainly stooping of the head and shoulders or excessive relaxation of body muscles.

• Points of interest:• -Shoulders• -Head• -Spine• -Hips

Source:http://www.merriam-webster.com/dictionary/slouch

SLOUCHING

• Tracking Points to be used:• Shoulder_Right• Shoulder_Left• Shoulder_Center• Head• Spine• Hip_Center• Hip_Right• Hip_Left

SLOUCHING

• Left Slouching:• Left shoulder will be

lower then the right shoulder.

• All points on left arm will be lower then base image.

• Head will be tilted to left.

• Left hip will be lower then right hip.

• Spine point will move slightly up and right.

Highlight what is slouchingBreak it into left right

=Movement Up =Movement Down

SOLUTIONS TO THE CHALLENGES OF DEFINING

• A multi point approach can help solve the majority of the problems when describing what is slouching.

• Use a combination of how much each point moves to determine if the subject is slouching or just moving one part of their body.

HEAD DISPLACEMENT

• Definition of Head Movement:• Voluntary or involuntary motion of head

that may be relative to or independent of body.

• http://www.medical-dictionary.cc/what-does/head-movement-mean

HEAD DISPLACEMENT

• Critical Tracking Points (TPs):

1. Head (H)

2. Shoulder_Center (SC)

• Associated Tracking Points (TPs):

1. Shoulder_Right (SR)

2. Shoulder_Left (SL)

DEFINITION OF HEAD MOVEMENT BASED ON TRACKING POINTSHead Movements Tracking Points Definition Changes in Coordinates

Lowering head H approaches SC X, Y, maybe Z too

Nodding H moves back and forth from SC repeatedly X, Y, maybe Z too

Head turning Turn to the left: H moved to the left X, Y, Z

Turn to the right: H moved to the right

Head tilted to one side Tilt to the left: H moved to the left X, Y

Tilt to the right: H moved to the right

Head bobbing H moves in random direction with minimal distance X, Y, Z

Head sliding forward H moves forward Z

Head wagging H moves in left and right rapidly X, Y

Source: http://www.thefreedictionary.com/Head+Movements

HOW WILL THIS WORK?A QUICK ROADMAP

Observation of error

Automatic identification

and classification of

error

Feedback to the user,

customized to their interaction

error

ACTIVITIES

• Link behavior and interaction to the image

• Understand the basic performance characteristics

• Relay back whether the interaction (or change in interaction) affects performance

The benefit is to examine the information associated with the sample, but also the video interaction of the image.

HBSI V3 has (will have):• video and audio

interaction– Watch the interaction– Understand who is

contributing the error– Replay the interaction in

real time as it was collected

• Metadata collected and searchable

QUESTIONS?

• Other actors• Contribution of the operator to the error• Contribution of the test administrator to the error

• HBSI workflow• Semi-automatic coding of the model using Kinect

• New work • Accessibility study – hearing and sight impaired (started Jan 2012)

• Contribution of cost to the model (started Jan 2011)• Examining the role of the impostor (thinking …. As this model only

has been rested in a “genuine” environment)• Development of products that can help improve interactions

Get Involved in shaping these projects – contact elliott@purdue.edu to participate in the development of the model

Teleconferences over the summer 2012 period